Data Type Conversion - 2022.1 English

Vitis Model Composer User Guide (UG1483)

Document ID
UG1483
Release Date
2022-05-26
Version
2022.1 English

Convert the input to the data type of the output.

The block warns or errors out when an integer or fixed-point output overflows during simulation. To configure, select Configuration Parameters > Diagnostics > Data Validity. In the Data Validity pane, set Wrap or Saturate to Overflow.

Library

Signal Attributes

Description

The Data Type Conversion block has one input and one output. It converts the value of the input signal to the data type of the output. This conversion tries to preserve the mathematical value of the input signal. The data type of the output is specified via the mask dialog. The conversion is governed by the following rules:

  • Conversions where the output data type is fixed-point, first select the nearest number that can be represented, taking into account the overflow mode. In case of a tie, the rounding mode breaks the tie.
  • Conversions where the output data type is integer are performed as in the C language. Overflow is handled via truncation.
    • As per IEEE Standard for Floating-point Arithmetic (IEEE Standard 754, Section-7.2), conversion from floating-point to integral is an invalid operation, when the floating-point value is outside the range of the destination integer data type. In this case, the output integer value depends upon the implementation of a C compiler. Hence, the results from the HLS Data Type Conversion block may differ from the results from Simulink┬« Data Type Conversion block.
    • When the floating-point input value is outside the range of the integer data type, the simulation results between Model Composer and RTL co-simulation in Vitis HLS may also differ.
    • During simulation, to check whether the input floating-point value goes outside the range of the destination integer type, in Simulink select Model Configuration Parameters > Diagnostics > Data Validity . Then set Saturate on overflow to either warning or error.
  • Conversions where the output data type is floating point follow the rules implemented in the C language.

Data Type Support

The input signal can be double, single, an integer, boolean, Xilinx supported half or Xilinx supported fixed-point data type.

The data type of the output is specified the mask parameters.

The input can be real or complex, and scalar, vector, or matrix. The output signal has the same complexity and dimensions as the input signal.

Parameters

Output data type

This parameter specifies the data type of the output signal. If fixed is specified, more parameters are available.

Settings for the Output data type parameter are as follows.

Table 1. Output Data Type Parameter
Setting Description
double double precision floating point
single single precision floating point
int8 8-bit signed integer
uint8 8-bit unsigned integer
int16 16-bit signed integer
uint16 16-bit unsigned integer
int32 32-bit signed integer
uint32 32-bit unsigned integer
logical boolean
fixed Xilinx supported fixed-point
half Xilinx supported half precision floating point
data type expression

A string that specifies the output data type. See "Working with Data Type Expression" in the Vitis Model Composer User Guide (UG1483).

Signedness

If the Output data type is set to fixed, the Signedness parameter specifies whether the output is a signed fixed-point or unsigned fixed-point data type.

Settings for the Signedness parameter are as follows.

Table 2. Signedness Parameter
Setting Description
Signed The output type contains both positive and negative numbers.
Unsigned The output type contains only non-negative numbers.

This parameter is available only if fixed is selected as the setting for parameter Output data type.

Word length

If the Output data type is set to fixed, the Word length parameter specifies the number of bits used to represent it.

Table 3. Word Length Parameter
Choices Description
16  
N A positive integer

This parameter is available only if fixed is selected as the setting for parameter Output data type.

Fractional length

If the Output data type is set to fixed, the Fractional length parameter specifies the number of bits to the right of the binary point.

Table 4. Fractional Length Parameter
Choices Description
10  
N An integer

This parameter is available only if fixed is selected as the setting for parameter Output data type.

Round

If the Output data type is set to fixed, the Round parameter allows you to select among five rounding and two truncation options.

If one of the five rounding options is selected, the block always rounds to the nearest supported precision. The five rounding choices are relevant only in case of a tie. For example, assume the output type is signed fixed-point, with a word length of 6 and a fractional length of 2, and the input to the block is 2.74. In this case, the output is rounded to the nearest supported precision, 2.75, regardless of which one of the five rounding modes is selected. However, if the input value is 2.625 (halfway between 2.5 and 2.75), then the output value depends on the chosen rounding mode. If Round to plus infinity is selected, the value will be 2.75, and if Round to zero is selected, the value will be 2.5. For more information on this, refer to the Vitis High-Level Synthesis User Guide (UG1399).

If one of the two truncation options is selected, the output will be truncated to the supported precision specified by the truncation selection.

Truncation to minus infinity is the default setting for Round and requires the smallest hardware resources among all the options.

The Round parameter is available only if fixed is selected as the setting for the Output data type parameter.

Settings for the Round parameter are:

Table 5. Round Parameter
Setting Description
Round to plus infinity Rounding to plus infinity
Round to zero Rounding to zero
Round to minus infinity Rounding to minus infinity
Round to infinity Rounding to infinity
Convergent rounding Convergent rounding
Truncation to minus infinity Truncation to minus infinity
Truncation to zero Truncation to zero
Overflow

If the Output data type is set to fixed, the Overflow parameter specifies the overflow mode applied during conversion.

This parameter is available only if fixed is selected as the setting for parameter Output data type.

Settings for the Overflow parameter are:

Table 6. Overflow Parameter
Setting Description
Saturation Saturation
Saturation to Zero Saturation to zero
Symmetrical Saturation Symmetrical saturation
Wrap around Wrap around
Sign-Magnitude Wrap Around Sign magnitude wrap around
Type Expression

If the Output data type is set to data type expression, the Type Expression parameter specifies the output data type as a string.

This parameter is available only if data type expression is selected as the setting for parameter Output data type.

Saturate on integer overflow

This parameter specifies whether integer overflow is handled by wrapping (default) or by saturating. This parameter is relevant only if the output is integral (int8, int16, int32, uint8, uint16, uint32).

When overflow is detected, the Diagnostic Viewer displays messages depending on the diagnostic action selected in the Configuration Parameters dialog box. To configure, in the Configuration Parameters > Diagnostics > Data Validity pane, set the Wrap or Saturate on overflow.

Settings for the Saturate on integer overflow parameter are:

Table 7. Saturate On Integer Overflow Parameter
Setting Description
Not selected Integer overflow is handled by wrapping.
Selected Integer overflow is handled by saturation.

If the Output data type is set to fixed and overflow is detected, the Diagnostic Viewer displays messages that depend on the diagnostic action you specify in the Simulink Editor. To configure, select Simulation > Model Configuration Parameters > Diagnostics > Data Validity for your model in the Simulink Editor, then set the Wrap on overflow or Saturate on overflow parameter.